Site Overlay

Limitations and Sources of Inaccuracy

This model is just that, a model. Models attempt to represent reality but will always be an approximation of what they aim to represent. Some of these approximations are significant.

  • This model is based on what epidemiologists call a simple SIR approach to the rate of new infection in an epidemic; Wikipedia describes some of these.
  • The approach is deterministic, which means that many things which vary randomly are taken to be fixed; an example is the length of time between stages of the illness and therefore the length of time an individual was infected before they died. This approach leads to a spurious degree of apparent accuracy in the output. A stochastic approach which takes account of randomness in the inputs would enable some indication of the range of uncertainty in the results to be given, but such an approach is beyond the intention and scope of this model.
  • Other significant inputs may not be able to be determined accurately, such as the percentage of those infected who show no symptoms.
  • The approach uses a simple graphical approach to determining growth rates (growth factors), which is believed to be a novel (or little used) approach for a free and widely available epidemic model of this type.
  • Central to this approach is the use of public data and this data itself may (and probably does) have inaccuracies for the purpose of the model, examples being that it may have been inaccurately recorded or that it is in some way unrepresentative of the underlying population being studied.
  • Implicit in the model’s construction is the assumption that, in the absence of any change, the appropriate rate of growth to use for projections into the future is that currently experienced as calculated by the model for the recent past, which may not be the case.
  • The internal calculations required to achieve the model’s apparent relative simplicity ar complex and written by one person. Many such spreadsheets contain errors – particularly difficult to detect and remove are errors which do not occur with all sets of input data. It is unlikely that all such errors have been removed especially as there has not yet been any peer review of the model.
  • Many epidemiological models are inaccurate with sparse early data and so is this and it becomes more accurate with more data.
  • The model also contains limitations as a result of the way the computer model is constructed or limitations within Excel; an example of the first is that the model does not in its current form deal with a longer period between infection and death than 30 days – and an example of the latter is how many days the graphs can easily project ahead; both of these limitations can be addressed but this has not yet been done with the current model.

These are some of the limitations of the model: some are shared with other epidemiological models but some are unique to this model. There may be, and probably are, other limitations and users should seek appropriate advice before using the output from the model.

At this time a paper on the model has just been submitted for publication on medRxiv. This is not a peer-reviewed journal. It is intended to submit the paper for publication in a peer-reviewed journal soon and the model should not be used until a paper has been published in a peer-reviewed journal.